The Rocky Mountain poll, conducted by the Behavior Research Center, is similar to the Franklin & Marshall poll in Pennsylvania. It doesn't really push the leaners, but in general is really accurate:
Let's analyze...
Their RV poll had Brewer up by 3, LV poll was Brewer up by 11, which should probably say something about screening process. It was conducted from October 1-10, so it is, at least somewhat fresh.
The poll for Senate had McCain up 28, he won by 24. The poll for Attorney General was once again tied among RV, with Horne up by 4 in the LV poll. Horne won by 4.
That poll is from May 2008. Which is way too old to be using in judging a polling company's performance.
Even though you didn't bring up 2006, their last poll was equally ancient, July 2006.
Conducted October 5-October 14, so mildly fresh, at least, like 2010.
You missed 2002, where they conducted a weekly poll of the Governor's race every week up to the election. The last iteration had Napolitano up by 9 among RV and 10 among LV (ok...). She won by 1 point, though this was a strong third-party race, and therefore more difficult to poll.
Only 1 point lead among RV, the 3-point was among LV. Conducted between October 4 and 6, so somewhat fresh, as before.